From the first token to the final footnote.
Open-source Markdown rendering for AI applications.
Incremental streaming, rich content, and thoughtful multilingual typography.
Source · Documentation · Migration guide · Discussions
AI output is a document in motion: a code fence may still be open, a formula unfinished, or a footnote definition yet to arrive. ai-markdown brings parsing, document coordination, and framework rendering together to handle those intermediate states.
- Streaming, with continuity. Incremental parsing where safe, reusable rendered blocks, and smooth text reveal in the React adapter.
- Rich answers. GFM tables and task lists, KaTeX math, and optional Mantine presentation for highlighted code and Mermaid diagrams.
- Multilingual by design. CJK-aware delimiter parsing and optional spacing between CJK and Latin text.
- Room for your UI. Typed component overrides, typography tokens, and configurable sanitization and URL policies.
| Package | What it brings |
|---|---|
| @ai-markdown/react | React components, hooks, streaming, and document coordination. Start here for React apps. |
| @ai-markdown/react-mantine | Mantine typography, code controls, syntax highlighting, and diagrams. |
| @ai-markdown/core | Framework-independent sessions, block plans, and document contributions. |
| @ai-markdown/engine | Markdown parsing, transforms, and incremental processing. |
| @ai-markdown/remark-mark-highlight | A standalone remark plugin for ==highlight== syntax. |
Release status: the shared core, engine, React, and Mantine packages are on the v3 beta train. Use the explicit @beta tag and follow the package README for peer dependencies and styles. The highlight plugin has its own release line. Vue support is in active development.
New to the project? Start with the React installation guide. Moving from @ai-react-markdown? Follow the package migration guide — the old React core package is now @ai-markdown/react; the new @ai-markdown/core is the shared runtime.
Streaming & performance · Smooth streaming · Document coordination · CJK typography · Architecture
Questions and ideas belong in Discussions. For bugs, open an issue with the package version, a minimal Markdown sample, and the exact sequence of streaming updates when relevant.
Ready to contribute? Read the contributor guide and code of conduct. For vulnerabilities, follow the security policy.
About the other repositories
The rehype-raw, hast-util-raw, and hast-util-from-parse5 repositories are forks of upstream unified ecosystem utilities. They support dependency maintenance; the main project, packages, and documentation live in ai-markdown/ai-markdown.
Markdown rendering for AI applications · Streaming · Math · CJK typography
Open source · MIT licensed · Built with the unified ecosystem
